[ntp:questions] Re: The asymmetry strikes again!
Richard B. Gilbert
rgilbert88 at comcast.net
Wed Jul 20 23:50:46 UTC 2005
Matt Kinard wrote:
>I'm still researching NTP issues over an asymmetric network, and I was
>wondering if I could get some suggestions as to what sort of accuracy
>I'm looking at sacrificing.
>
>Are we talking 20 ms or are we talking about 1 or 2 seconds? Is it
>contingent upon how asymmetric my network is? What is the threshold?
>
>For the particular application I'm studying, there is some acceptable
>risk in terms of lost accuracy, but I need some more concrete ideas.
>
>Thanks all!
>Matt
>
>
>
>
The error in transmitting time from server to client via NTP is, at
most, one half the round trip delay. That's about all you can say
about it except that typically the error is less than that.
This is a good reason to try to find servers close to you in network
space; e.g. with low round trip delays.
Note that the error we are talking about here is only the portion of the
total error due to the process of transmitting the time from server to
client. Monitoring internet servers while using a stable and accurate
reference such as a GPS timing receiver will show that the quality of
these servers is highly variable. Some are stable as a table and very
close to the time from the GPS receiver. Others tend to wander plus or
minus twenty or thirty milliseconds during the course of a day. This
is why you should configure four to seven internet severs if you are
using the internet as your sole source of time. The quality of the
network is also highly variable; round trip delays tend to be greatest
during business hours on weekdays.
More information about the questions
mailing list